Comparing approaches to predict transmembrane domains in protein sequences

Abstract

There are today several systems for predicting transmembrane domains in membrane protein sequences. As they are based on different classifiers as well as different pre-and post-processing techniques, it is very difficult to evaluate the performance of the particular classifier used. We have developed a system called MemMiC for predicting transmembrane domains in protein sequences with the possibility to choose between different approaches to pre-and post-processing as well as different classifiers. Therefore it is possible to compare the performance of each classifier in a certain environment as well as the different approaches to pre-and post-processing. We have demonstrated the usefulness of MemMiC in a set of experiments, which shows, e.g., that the performance of a classifier is very dependent on which pre-and post-processing techniques are used.